Decentralized and Dynamic Zero-Touch Provisioning of Leaf-Spine EVPN Data Centers

Data centers are becoming increasingly large and complex due to the exponential growth of data and the rising demand for cloud services. To manage this complexity, modern data center networks employ leaf-spine topologies and EVPN overlays, enabling efficient scaling and supporting multi-tenancy. These networks often include thousands of servers and network devices, making configuration and management highly challenging. This paper proposes a decentralized and dynamic Zero-Touch Provisioning (ZTP) solution for leaf-spine EVPN data centers. Our approach employs autonomous agents on each data center router to provision both the underlay and the overlay infrastructure without manual intervention, handling both initial deployments and disruptive events (e.g., router or link failures). The underlay is assumed to use a link-state routing protocol, and we introduce a novel algorithm that infers the role of each router (e.g., leaf or spine) from the network topology provided by the protocol, a feature central to our approach. We experimentally validated our solution using SR Linux, a network operating system for data centers that supports non-native agent integration. Our results demonstrate that the proposed ZTP solution can fully provision leaf-spine data center networks automatically while achieving very fast convergence times.